MCP servers are kind of the big thing right now. We have worked on RAG models in the past and found them both fun and incredibly useful! When we learned about MCP servers and their promise to handily increase the ability of AI assistants, we were excited to give this new technology a try!
Text-to-Website takes a user's request for a website (or other coding tasks) in plain English, takes a few minutes to freshen up, and then gives you a link to the website you asked for, along with custom-made documentation to help you edit and maintain the website for yourself!
How we built it
We're using a standard io MCP server to expose system tools to the Google Gemini API to allow it to directly access our file system to run commands and generate code.
Originally, we were going to run Google's Gemma3n model locally, but it didn't play very nicely with any of our MCP tools, so we had to pivot. On the bright side, Gemini is wayyy faster than Gemma for generating code!
Not only did we create the project we had more or less set out to make, but it blew us away with its abilities! Not only can it generate complex and visually stunning websites in minutes, but it's also capable of more complex, multi-file projects (with varying success)!
We learned so much during these past three days, it would probably be easier to talk about what we haven't learned yet! We spent about 10 hours messing about with Gemma, Ollama, LM Studio, and MCP tools that didn't want to work. The frontend was a React application until about 10 hours before we had to turn something in (it ended up being a Tkinter Python UI)!
We're both incredibly proud of what we've made and hope to further develop it, giving it heightened abilities, such as a more complex UI and WAN website hosting.
Built With
- gpt5
- python
- tkinter
- vscode




Log in or sign up for Devpost to join the conversation.